Coca-Cola's AI Demand Forecasting Boosts Sales by 8%

Neeraj Tolmare, the chief information officer at Coca-Cola, emphasizes that the artificial intelligence pilots he invests in must demonstrate potential for revenue generation or significant efficiency gains once fully implemented. This approach aligns with a directive from Coca-Cola's CEO James Quincey and the broader leadership team, which mandates that AI experimentation must be tied to tangible outcomes that drive meaningful results for the company.

Coca-Cola, ranked 97th on the Fortune 500, operates on a massive scale, serving 2.2 billion beverages daily. Each AI initiative by Tolmare must consider the cost of implementation and how the technology can be utilized across various functions, from software development to sales. Additionally, Tolmare must evaluate how advancements like agentic AI will impact workflows and data sharing within Coca-Cola and with key external partners, including 200 bottlers and 950 production facilities.

One notable pilot project involves an algorithm developed by Coca-Cola to help retail outlets better predict demand. This AI-driven solution uses historical data, weather patterns, and geolocation data from Google to more accurately forecast future sales. The insights from this algorithm inform messages sent via WhatsApp to managers, advising them on when to restock products like Sprite or Diet Coke. This pilot, tested in three countries, resulted in a 7% to 8% increase in sales compared to outlets not using the AI algorithm. Tolmare plans to expand this AI tool to more markets globally.

Another significant investment is Coca-Cola's use of AI for content creation. The company sells in over 180 countries and creates marketing assets in more than 130 languages, a process that is both time-consuming and costly. Coca-Cola recently created 20 AI-generated assets based on its proprietary intellectual property and produced 10,000 variations for use across different languages and geographies. Tolmare notes that consumers are 20% more likely to engage with this AI-produced content, which was generated three times faster than previous iterations.

Tolmare stresses the importance of keeping humans involved in the AI process. While AI can unlock value that humans cannot, due to the complexity of data mining, its application in marketing requires careful consideration. An AI-generated Christmas spot and a campaign featuring a misattributed quote from a book by J.G. Ballard highlighted the challenges of using AI for creative content. Coca-Cola has strict guidelines to ensure that AI-generated content avoids social biases and is aware of potential issues like hallucinations or deepfakes.

Tolmare's career spans over two decades in the technology industry, including roles at Palm, Cisco Systems, and HP. He joined Coca-Cola in 2018 to leverage technology for business growth rather than developing it. Tolmare spearheaded Coca-Cola's transition to cloud computing, retiring or selling all of its physical data centers. The company's cloud footprint is primarily on Microsoft Azure, with the rest split between Amazon Web Services and Google Cloud. Tolmare continuously monitors which cloud provider offers better efficiency for moving workloads.

Coca-Cola's hybrid cloud strategy is also applied to its generative AI experimentation. The company collaborates closely with Microsoft and OpenAI but also works with other AI hyperscalers like Google, Meta, and Anthropic. This approach ensures that Coca-Cola is not locked into a single provider too early, as the market is still evolving.

Looking ahead, Coca-Cola is exploring agentic AI, systems designed to perform tasks autonomously or with minimal human intervention. The company is considering offerings from vendors like Microsoft, SAP, and Adobe, while developing its own AI agents trained on Coca-Cola's data. All these AI agents are still in the pilot phase as Tolmare determines their cost-effectiveness and optimal business outcomes.

Tolmare is enthusiastic about the potential of agentic AI for Coca-Cola's business, stating that while the technology has not yet been launched in production, the company is very close to doing so.